How to view the future of OpenAI?
By Wang Ziwei @ Retail Wei Observation
On September 26th, the Wall Street Journal reported that OpenAI is in negotiations with investors regarding the possibility of selling stocks based on the company's valuation. This would allow employees to cash out by selling stocks to external investors.
At the same time, OpenAI's valuation has tripled from around $27 billion in March of this year to $80 to $90 billion. In terms of revenue, sources familiar with the matter quoted by the Wall Street Journal stated that OpenAI is expected to generate $1 billion in revenue this year and reach billions of dollars by 2024.
With a valuation of $90 billion and revenue of $1 billion, OpenAI's price-to-sales ratio (P/S) reaches an astonishing 90 times, far exceeding that of tech giants like Microsoft and Nvidia.
Let's analyze what kind of story OpenAI will tell in order to achieve a reasonable valuation of $90 billion while rapidly increasing its revenue. Our analysis will be conducted from two perspectives: 2B and 2C.
[One] 2B: The Birth of "Model Factory" for Large Language Models#
Large language models (LLMs) have continuously broken through the limitations of scale as computer computing power exponentially grows, and the emergence of GPT-3 is a typical example.
Currently, the scaling law of models is still being pushed forward, and it is foreseeable that larger-scale LLMs will appear in the future, possessing stronger language understanding and generation capabilities than smaller models.
However, training top-level LLMs is extremely difficult. Looking at global companies, only a few can achieve this. Moreover, even if the model parameters are obtained, there is a lack of professional knowledge to start pre-training from scratch. These two points are precisely the core problems that ordinary companies face in the era of AI large models.
Therefore, the development of LLMs must be outsourced to "model factories" like OpenAI: With the powerful ability to train top-level models, OpenAI can derive and design thousands of small customized models to meet the personalized needs of different customers. This is similar to Apple designing custom A-series chips for different models of iPhones.
The current biggest obstacle to using large models like GPT is cost (the second is data compliance and privacy). However, as new models like GPT-4 continue to be developed, hardware costs will decrease due to economies of scale, and usage costs will also decrease significantly. The innovation and optimization of OpenAI's "model factory" in the future will make advanced language models accessible to the general public.
For OpenAI, the profitability of the 2B model depends on the benefits brought by the improvement of customer productivity, and a certain "technology usage fee" can be charged. This "Model-as-a-Service" model will also pose a challenge to traditional SaaS companies.
[Two] 2C: Subscription is Just the First Step, the Future is "Super Entry"#
In March of this year, The Information, a technology website, reported that 70-80% of OpenAI's revenue comes from 2C subscriptions, specifically the $20/month Chat GPT Plus, with a subscriber base of approximately 2 million at the time.
It must be pointed out that any new technology may be worthless until we find a "killer application." Jeff Bezos, the founder of Amazon, said in a speech in the early years that electricity was difficult to enter ordinary households because people didn't know what electricity was for until someone invented the light bulb, and then electricity truly became infrastructure.
Later, Bezos also said in an interview that invention and creation are not disruptive, but acquiring users is the real disruption. Chat GPT can be seen as such a case, and Chat GPT Plus is also a recognition of GPT by users.
Therefore, a $20 monthly subscription fee can at most be a trial to make everyone realize that users are willing to pay for this application, and this application is a killer application.
In fact, the real vast ocean for 2C is to become a "super entry" on mobile devices.
Imagine picking up your phone, unlocking it, and there is only one OpenAI intelligent assistant app on the home screen—just like when the iPhone removed the keyboard and left only a "Home" button.
You say to the intelligent assistant, "Help me plan a trip to Japan," and the intelligent assistant can proactively contact services such as flights, hotels, and attractions, and provide you with a detailed itinerary that includes all the details. It can also adjust the details according to your requirements. OpenAI may only need a few minutes to generate a very comprehensive plan for you, which means you don't need to personally open apps like Xiaohongshu and Mafengwo to look for travel guides, or open Ctrip and Fliggy to compare flight prices. The "intelligence" built by many apps on your phone will instantly crumble.
As for whether this intelligent assistant will talk nonsense, let's put it this way: On the one hand, OpenAI can already be connected to the internet, and on the other hand, even in the era before internet connectivity, a friend who specializes in custom African tours told me that their tests showed that under the same customer requirements, the solutions provided by Chat GPT are not worse than those of travel planners with 7-8 years of experience.
As for price comparison, that can be easily solved. Just refer to Amazon's logic: Amazon uses dynamic pricing to ensure that the prices of products on its website are lower than those of competitors. Once consumers become familiar with this, they will no longer compare prices but directly "one-click order" on Amazon—what they don't know is that Amazon is no longer the cheapest at this point.
For OpenAI, this 2C super entry model can strongly challenge the revenue of mobile app stores. The global revenue of smartphone app stores has already reached $50 billion, but this is just the beginning. With the emergence of super entry, the $450 billion global digital advertising revenue may be redistributed.
[Three] Conclusion#
Whether it is the 2B "model factory" or the 2C "super entry," OpenAI has the potential to ride the waves and enter a trillion-dollar race. The 2B and 2C tracks themselves will also generate synergy. Even if only one model ultimately succeeds, achieving a revenue target of billions of dollars is a small problem.
In the future, whether these stories can be told depends on whether the scaling law of models continues to advance. As long as the three elements—computing power, data, and algorithms—continue to make progress, LLMs will inevitably improve, and costs will decrease. If all obstacles can be overcome, OpenAI's valuation is not a fantasy, and that is what we can look forward to.
As Microsoft founder Bill Gates said, most of us overestimate what we can do in one year and underestimate what we can do in five years.
"Retail Wei Observation" focuses on the latest strategies, tactics, and thoughts in the field of new retail and new consumption from a global perspective. The founder of the platform, Wang Ziwei, is an independent analyst.